Hand shape classification is an important problem for the human computer interaction and the fingerspelling recognition. These problems need a real-time process and scale invariance. To this end, we propose a feature vector for hand shape classification which is fast, and robust to scale. The proposed method calculates an adaptive k-curvature which computes a curvature depending on the hand size. The proposed method works at 0.08sec per image and has a 99% accuracy.